Is TikTok's algorithm like a digital newspaper editor?

No, unlike a newspaper editor who curates based on values, TikTok's algorithm is a machine learning system that builds mathematical approximations to predict user preferences, making it agnostic to human values.

Why is TikTok so effective at recommending videos?

Its success stems from the short-form video format which provides frequent user feedback, and its powerful distributed system that can update user profiles in near real-time, combined with mixing in trending content.

Are machine learning algorithms ethical?

Machine learning algorithms are inherently amoral; they optimize for goals without understanding human values or ethics. This can lead to the amplification of harmful or dark human impulses.

Can changing TikTok's operators in the US 'fix' the algorithm's impact?

No, the problem is baked into the mathematical nature of recommendation systems, not the nationality or specific values of the people controlling them. Algorithms lack inherent values.

Should I stop using social media if I use it minimally?

Focus on whether the technology harms your values or life quality. If minimal use doesn't degrade your life or civic values, it might be acceptable, but be mindful of contributing to the ecosystem.

Is TikTok a good way for teens to stay informed about the news?

TikTok is a terrible way for teens to get news because its algorithmic curation lacks human oversight and can amplify misinformation or exploit darker impulses, rather than focusing on truth or civic value.

What are better alternatives for staying informed than TikTok?

Seek out human-curated news sources like various newspapers, aggregated news sites (e.g., AllSides), podcasts, or newsletters that involve human judgment and values.

Can quantum computing solve the scaling issues in AI?

Quantum computers are not a universal upgrade for current computing tasks and are limited to specific problem types. They are unlikely to solve the scaling barriers for general generative AI models in the near future.

How can I prevent my life from being dominated by algorithmic curation?

Make your life outside of your phone deep and engaging. When your real life is rich and fulfilling, digital diversions like TikTok become less compelling.

What's the role of parents in kids' social media use?

Parents should act as the primary filter and interpreter of the world for their children, especially during formative years, rather than letting algorithms dictate values and exposure to online content.

What are the real-world negative impacts of TikTok's algorithm?

Recent articles show inadequate child data protection, the amplification of end-of-the-world beliefs through 'Rapture Talk,' and the promotion of dangerous stunts that have led to tragedies.

How Does TikTok Actually Work? (It’s Scary…)

Deep Questions with Cal Newport

People & Blogs4 min read86 min video

Sep 29, 2025|11,345 views|276|39

Cal Newport Deep Work Deep Life Deep Questions TimblockPlanner Deep Questions Podcast cal newport interview cal newport podcast social media detox productivity tips cal newport productivity cal newport motivation

Save to Pod

Key Moments

TL;DR

TikTok's algorithm is a complex "two-tower" system, not an "editor." It blindly optimizes for engagement, potentially exploiting dark human impulses. US control won't fix its fundamental nature.

Key Insights

TikTok's "algorithm" is actually a sophisticated "two-tower" recommender system architecture, not a simple editorial process.

This system uses machine learning to create embeddings (lists of numbers) for both users and videos, matching them based on learned patterns.

The system's "intelligence" comes from its ability to process vast amounts of user feedback (watch time, likes, swipes) in near real-time, which short-form video is ideal for.

Recommendation algorithms are inherently value-agnostic, optimizing for engagement by approximating underlying systems, which can include undesirable human impulses.

Changing the ownership of TikTok to US control (e.g., Oracle) is unlikely to fundamentally fix the system's core issues, as the algorithm itself lacks human values.

Human curation, despite potential biases, is preferable to algorithmic curation for news and information, as it incorporates human values and guardrails.

THE ALGORITHM DEBATE: FROM MYTH TO REALITY

The recent news surrounding potential US control of TikTok's operations centers on its powerful recommendation algorithm. Many perceive this algorithm as a digital editor, shaping content for users. However, this understanding is flawed. The discussion highlights a common misconception where the term "algorithm" is imbued with almost mystical power. Cal Newport, drawing on his computer science background, aims to demystify how these systems actually function, moving beyond the popular narrative to reveal the underlying technical architecture. The goal is to understand the system's true capabilities and limitations, especially in relation to its societal impact.

THE TECHNOLOGY BEHIND THE FEED: A "TWO-TOWER" SYSTEM

At its core, TikTok's recommendation engine is not a single algorithm but a complex distributed system, more accurately termed a "recommender system architecture." This system likely employs a "two-tower" model. One tower processes incoming videos, generating a "property list" or vector representing each video's characteristics. The second tower processes user profiles, particularly their on-platform behavior, to create a similar vector representing their interests. Both towers use machine learning, including neural networks and transformers, to map videos and users into a common vector space.

MACHINE LEARNING AND REAL-TIME FEEDBACK LOOPS

The "training" of these towers is largely semi-supervised, driven by vast amounts of user interaction data. The system learns by observing which videos users watch, like, or skip. The goal is to align the user's interest vector closely with the video's property vector. TikTok's success stems partly from the short-form video format, which generates rapid feedback. Users can cycle through dozens of videos per session, providing a rich, real-time data stream that allows the system to continuously update user profiles, leading to remarkably accurate initial recommendations and a strong "cold start" capability.

BEYOND PURE USER PREFERENCE: POPULARITY AND DIVERSITY

While user-specific interests are primary, TikTok's system also incorporates elements of trending or globally popular content. This "short-term profile" complements the "long-term profile" (user description). By mixing highly personalized recommendations with broadly popular or trending videos, the system introduces novelty and discovery. This strategy allows users to encounter content that might not precisely match their past behavior but is currently resonating with a wider audience. This blend contributes to the addictive and surprising nature of the feed, preventing it from becoming too predictable.

THE ETHICAL DUALITY: BLIND OPTIMIZATION VERSUS HUMAN VALUES

Crucially, machine learning recommendation systems like TikTok's are designed to optimize for engagement metrics, not to embody human values. They are essentially building mathematical approximations of underlying patterns in the data. This process is inherently value-agnostic. The system doesn't understand concepts like truth, ethics, or harm; it merely seeks to maximize positive feedback signals. As a result, it can inadvertently amplify negative human impulses like hatred, violence, or sensationalism, much like a "psychopathic newspaper editor" that lacks empathy and humanistic guardrails, leading to potentially harmful content proliferating.

RETHINKING CONTROL AND CURATION: HUMAN VS. ALGORITHMIC

The idea of US control over TikTok's algorithm, while addressing potential foreign influence, does not solve the fundamental problem. The algorithmic architecture itself is the issue. It operates blindly, optimizing for engagement without ethical consideration. This contrasts sharply with human curation, which, despite potential biases, incorporates normative standards and values developed over centuries. For news and information consumption, human-curated sources or those with human oversight are preferable, allowing for context, critical evaluation, and a consideration of civic values that algorithms inherently lack. The focus should shift from who controls the algorithm to the inherent limitations of algorithmic curation itself.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●Books

●Concepts

●People Referenced

Common Questions

TikTok uses a 'two tower' system where one tower processes videos to create property lists and the other processes user data to create interest profiles. These are then matched to recommend videos.

Topics

Algorithmic Curation Information Consumption

Mentioned in this video

conceptRSA public key encryption

Mentioned as a core method for secure commerce on the internet that relies on large prime factors, potentially vulnerable to quantum computing.

podcastTwo Dads, One Car

The podcast hosted by Stephen, who previously had Cal Newport as a guest.

bookEmma

Mentioned sarcastically as a potential positive TikTok influence, contrasting with negative real-world examples.

conceptGutenberg printing press

Referenced to highlight the historical context of mass content production and its impacts.

conceptUFO

Referenced as a theme for a Halloween haunt, with a specific mention of a 9-foot inflatable prop from Wayfair.

organizationMeta AI

Mentioned as publishing research papers about their work.

softwarereclaim.ai

A smart calendar assistant that defends focus time, autoschedules meetings, and resolves conflicts.

softwareMonarch Money

An all-in-one personal finance tool recommended for its visibility and ease of use.

Found this useful? Build your knowledge library

Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.

Try Summify free